|
--- |
|
license: apache-2.0 |
|
base_model: alignment-handbook/zephyr-7b-sft-qlora |
|
tags: |
|
- alignment-handbook |
|
- trl |
|
- dpo |
|
- generated_from_trainer |
|
- trl |
|
- dpo |
|
- generated_from_trainer |
|
datasets: |
|
- HuggingFaceH4/ultrafeedback_binarized |
|
model-index: |
|
- name: zephyr-7b-dpo-qlora-fsdp |
|
results: [] |
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# zephyr-7b-dpo-qlora-fsdp |
|
|
|
This model is a fine-tuned version of [alignment-handbook/zephyr-7b-sft-qlora](https://huggingface.co/alignment-handbook/zephyr-7b-sft-qlora) on the HuggingFaceH4/ultrafeedback_binarized dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.8742 |
|
- Rewards/chosen: 0.0082 |
|
- Rewards/rejected: 0.0003 |
|
- Rewards/accuracies: 0.6726 |
|
- Rewards/margins: 0.0079 |
|
- Logps/rejected: -242.3632 |
|
- Logps/chosen: -266.8597 |
|
- Logits/rejected: -2.3743 |
|
- Logits/chosen: -2.4108 |
|
|
|
## Model description |
|
|
|
More information needed |
|
|
|
## Intended uses & limitations |
|
|
|
More information needed |
|
|
|
## Training and evaluation data |
|
|
|
More information needed |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 5e-06 |
|
- train_batch_size: 10 |
|
- eval_batch_size: 8 |
|
- seed: 42 |
|
- distributed_type: multi-GPU |
|
- num_devices: 6 |
|
- gradient_accumulation_steps: 4 |
|
- total_train_batch_size: 240 |
|
- total_eval_batch_size: 48 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: cosine |
|
- lr_scheduler_warmup_ratio: 0.1 |
|
- num_epochs: 5 |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen | |
|
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:| |
|
| 0.2536 | 0.39 | 100 | 0.2792 | 0.0024 | 0.0006 | 0.6042 | 0.0019 | -242.3385 | -267.4340 | -2.3735 | -2.4122 | |
|
| 0.5352 | 0.79 | 200 | 0.5010 | 0.0011 | -0.0019 | 0.5744 | 0.0030 | -242.5832 | -267.5640 | -2.3629 | -2.4014 | |
|
| 0.3676 | 1.18 | 300 | 0.8293 | 0.0079 | 0.0027 | 0.5982 | 0.0052 | -242.1211 | -266.8856 | -2.3788 | -2.4168 | |
|
| 0.366 | 1.57 | 400 | 0.8239 | 0.0065 | 0.0007 | 0.6399 | 0.0058 | -242.3221 | -267.0256 | -2.3774 | -2.4146 | |
|
| 0.292 | 1.96 | 500 | 0.8146 | 0.0050 | -0.0005 | 0.6399 | 0.0055 | -242.4462 | -267.1794 | -2.3978 | -2.4343 | |
|
| 0.1355 | 2.36 | 600 | 0.9651 | 0.0047 | -0.0013 | 0.6161 | 0.0060 | -242.5212 | -267.2061 | -2.3796 | -2.4178 | |
|
| 0.1327 | 2.75 | 700 | 0.9985 | 0.0046 | -0.0019 | 0.6339 | 0.0065 | -242.5883 | -267.2230 | -2.3690 | -2.4066 | |
|
| 0.0389 | 3.14 | 800 | 0.8932 | 0.0080 | 0.0003 | 0.6518 | 0.0078 | -242.3696 | -266.8748 | -2.3563 | -2.3947 | |
|
| 0.029 | 3.53 | 900 | 0.9392 | 0.0090 | 0.0008 | 0.6577 | 0.0082 | -242.3114 | -266.7798 | -2.3752 | -2.4118 | |
|
| 0.0198 | 3.93 | 1000 | 0.8200 | 0.0087 | 0.0010 | 0.6577 | 0.0077 | -242.2917 | -266.8047 | -2.3780 | -2.4145 | |
|
| 0.0059 | 4.32 | 1100 | 0.8904 | 0.0080 | 0.0002 | 0.6577 | 0.0078 | -242.3739 | -266.8760 | -2.3744 | -2.4108 | |
|
| 0.0042 | 4.71 | 1200 | 0.8779 | 0.0080 | 0.0001 | 0.6518 | 0.0080 | -242.3892 | -266.8771 | -2.3753 | -2.4119 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.38.1 |
|
- Pytorch 2.2.0+cu118 |
|
- Datasets 2.17.1 |
|
- Tokenizers 0.15.2 |
|
|
